Interview dates: Sept 17-21, 2012

Base: 2,259 registered voters (RV)
Base for Voting Intention: 1,616 Likely Voters (LV)

Ipsos Poll conducted for Reuters
These are findings from an Ipsos poll conducted for Thomson Reuters from Sept 17-21, 2012. For the survey, a sample of
2,259American registered voters (age 18 and over) was interviewed online. The precision of the Reuters/Ipsos online polls is
measured using a credibility interval. In this case, the poll has a credibility interval of plus or minus 2.5 percentage points for
Registered Voters and 2.9 for Likely Voters. For more information about credibility intervals, please see the appendix.
The data were weighted to the U.S. current population data by gender, age, education, and ethnicity. Statistical margins of error
are not applicable to online polls. All sample surveys and polls may be subject to other sources of error, including, but not limited
to coverage error and measurement error. Figures marked by an asterisk (*) indicate a percentage value of greater than zero
but less than one half of a per cent. Where figures do not sum to 100, this is due to the effects of rounding.


Q1. If the 2012 Presidential Election were being held today and the candidates were [ROTATE] Barack Obama for president and
Joe Biden for vice president, the Democrats, and Mitt Romney for president and Paul Ryan for vice president, the Republicans
[END ROTATE], for whom would you vote?

Voters (LV)
All Registered
Voters (RV)
Barack Obama for president and Joe Biden
for vice president, the Democrats
48% 48% 85% 9% 32%
Mitt Romney for president and Paul Ryan
for vice president, the Republicans
42% 39% 8% 80% 34%
Wouldn’t vote 1% 2% 1% 1% 5%
None / Other 3% 4% 2% 4% 12%
Don’t know / Refused 6% 8% 4% 6% 17%


Strong Democrat 19%
Moderate Democrat 21%
Lean Democrat 7%
Lean Republican 8%
Moderate Republican 16%
Strong Republican 14%
Independent 11%
None of these 2%
DK 2%

1146 19
St., NW, Suite 200
Washington, DC 20036
(202) 463-7300

How to Calculate Bayesian Credibility Intervals

The calculation of credibility intervals assumes that Y has a binomial distribution conditioned on the parameter θ\, i.e.,
Y|θ~Bin(n,θ), where n is the size of our sample. In this setting, Y counts the number of “yes”, or “1”, observed in the sample, so
that the sample mean (y ̅) is a natural estimate of the true population proportion θ. This model is often called the likelihood
function, and it is a standard concept in both the Bayesian and the Classical framework. The Bayesian
statistics combines both
the prior distribution and the likelihood function to create a posterior distribution. The posterior distribution represents our
opinion about which are the plausible values for θ adjusted after observing the sample data. In reality, the posterior
distribution is one’s knowledge base updated using the latest survey information. For the prior and likelihood functions
specified here, the posterior distribution is also a beta distribution (π(θ/y)~β(y+a,n-y+b)), but with updated hyper-parameters.

Our credibility interval for θ is based on this posterior distribution. As mentioned above, these intervals represent our belief
about which are the most plausible values for θ given our updated knowledge base. There are different ways to calculate these
intervals based on n(0¡˳). Since we want only one measure of precision for all variables in the survey, analogous to what is
done within the Classical framework, we will compute the largest possible credibility interval for any observed sample. The
worst case occurs when we assume that a=1 and b=1 and ˳ = n 2 / . Using a simple approximation of the posterior by the
normal distribution, the 95% credibility interval is given by, approximately:

˳ ∓


For this poll, the Bayesian Credibility Interval was adjusted using standard weighting design effect 1+L=1.3 to account for
complex weighting

Examples of credibility intervals for different base sizes are below.

Sample size Credibility intervals
2,000 2.5
1,500 2.9
1,000 3.5
750 4.1
500 5.0
350 6.0
200 7.9
100 11.2

Bayesian Data Analysis, Second Edition, Andrew Gelman, John B. Carlin, Hal S. Stern, Donald B. Rubin, Chapman & Hall/CRC |
ISBN: 158488388X | 2003

Kish, L. (1992). Weighting for unequal Pi . Journal of Official, Statistics, 8, 2, 183200.

Sign up to vote on this title
UsefulNot useful